JP2003250036A - Electronic watermark embedding method and electronic watermark detecting method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To match a position for detecting a signature even when geometrical processing such as reduction, magnification and rotation is applied to an image in addition to segmentation of the image. <P>SOLUTION: In a step 11, a predetermined object is detected from an image inputted in a step 10, and the area is segmented. Then a predetermined characteristic point is extracted from the area in a step 12, and a coordinate axis is set onto the image on the basis of the characteristic point in a step 13. In the case of extracting an object area from a received image in the step 11, the object area with many characteristic points such as the face of a person or the like is proper, and when the object area is the face, for example, eyes are made the characteristic points to set the coordinate axis. Then in a next step 14, a signature is embedded to the image on the basis of the coordinate axis set in the step 13 and the resulting image is outputted as the image to which the watermark is embedded in a step 15. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital watermark technique for concealing other information (signature) in image data for the purpose of image copyright management, and more particularly to a digital watermark embedding method for embedding a signature as a digital watermark. And a digital watermark detection method for detecting a signature as a digital watermark.

[0002]

2. Description of the Related Art FIG. 9 is a flow chart showing a digital watermarking method disclosed in WO99 / 01980, for example. In the figure, 90
Is a step of wavelet transforming an image, 91 is a step of spread spectrum of LL component (low frequency component of wavelet transform), 92 is a step of dividing spread spectrum data into blocks, and 93 is a step of Fourier transforming each block , 94 is the Fourier transform
This is a step of concealing 1 bit of the signature data in the DC component.

Next, the operation will be described. In step 90, only low frequency components (referred to as LL components) of the image signal are extracted. The signature data is embedded in this LL component. Step 91 multiplies the LL component by a predetermined random number sequence. This operation spreads the frequency so that the signature data cannot be easily detected. Next, this frequency-spread LL component is divided into small blocks in step 92. One bit of the signature data is embedded in each block divided here. Specifically, the information is concealed by Fourier transforming the block in step 93 and quantizing the DC component into a different representative value in step 94 depending on the signature bit. Although not shown, this is followed by the reverse steps to restore the image, creating a watermarked image. To detect the signature data from this image, the same steps can be repeated and the value of the DC component of the Fourier transform can be checked.

As described above, in the conventional digital watermark method, the original block needs to be correctly specified at the time of detection, and therefore, when a part of the image is cut, the original image is saved. It had to be compared with that in advance, or embedded special information for alignment, and used it for alignment.

In order to solve this problem, Japanese Patent Application Laid-Open No. 11-41453 discloses a digital watermark method in which an object is detected from an image and a coordinate system is determined on the basis of the object.

FIG. 10 is a flow chart showing the digital watermark system shown therein. In the figure, 100 is an original image, 1
01 is the step of extracting an object from this image,
102 is a step of obtaining a rectangular area circumscribing the object, 103 is a step of determining the position of a block in which the digital watermark information is embedded from the center point of the rectangular area, 104
Is a step of embedding a digital signature at the determined block position, and 105 is an image in which a watermark is embedded. Also,
On the detection side, 106 is a step of extracting an object from an image, 107 is a step of obtaining a rectangular area circumscribing the object, 108 is a step of determining the position of a block in which an electronic signature is embedded from the center point of the rectangular area, and 109 is This is a step of reading an electronic signature from the determined block position.

Next, the operation will be described. In this digital watermark method, a specific object is extracted from the image, a rectangular area circumscribing it is cut out, the coordinate axis with its center as the origin is found, and the position of the block in which the watermark is embedded is specified based on this coordinate axis. is doing. Therefore, even if a part of the image is cut out, the alignment for detecting the watermark is possible without requiring the original image or the like.

[0008]

However, JP-A-11-
Since the conventional digital watermark method disclosed in Japanese Patent No. 41453 is configured as described above, for example, as an attack for invalidating the digital watermark, a part of the image is cut off and the image is reduced. When geometrical processing other than clipping is performed, such as enlargement, enlargement, and rotation, it is difficult to align the watermark data, and it is difficult to specify the embedded position of the signature. Further, since the rectangular area including the object is cut out to specify the coordinate axis, the accuracy of the position is not sufficient, and there is a problem that the coordinate axis does not easily match at the time of embedding and at the time of detection.

The present invention has been made to solve the above problems, and not only cuts an image but also
Even if geometric processing such as reduction, enlargement, rotation, etc. is performed, the position for signature detection can be adjusted and the coordinate axes can be more accurately matched at the time of embedding the digital watermark and at the time of detection. It is an object of the present invention to provide a digital watermark embedding method and a digital scan detection method that can be performed.

[0010]

In order to solve the above problems, according to the present invention, an input step of inputting an image, an area extraction step of extracting an object area from the input image, and a feature point from the object area Extracting a feature point, a coordinate axis setting step for determining a coordinate axis based on the feature point in the input image, a signature embedding step for embedding a digital signature in the input image based on the coordinate axis, and the signature A digital watermark embedding method, which comprises an output step of outputting the generated image.

Further, an input step of inputting an image, an area extraction step of extracting an object area from the input image, a feature point extraction step of extracting a feature point from the object area, and an input using the feature point as a reference. A normalization step of normalizing the image, a coordinate axis setting step of defining a coordinate axis based on the feature points in the normalized image, and a signature embedding step of embedding a digital signature in the normalized image based on the coordinate axis And an inverse normalization step of performing an inverse transformation on the image in which the signature is embedded, a synthesis step of synthesizing the denormalized image with the original image, and outputting the synthesized image And a digital watermark embedding method.

The feature point extracting step is characterized by extracting a plurality of feature points.

Further, in the normalizing step, the size of the object is normalized.

Further, the normalizing step is characterized in that the angle of the object is normalized.

Further, the region extracting step is characterized by extracting a human face portion.

Further, an input step of inputting an image, an area extracting step of extracting an object area from the input image, a feature point extracting step of extracting a feature point from the object area, and a feature of the input image. The digital watermark detecting method is characterized by comprising a coordinate axis setting step of determining a coordinate axis based on a point and a signature detecting step of detecting a digital signature from an image input based on the coordinate axis.

Further, an input step of inputting an image, an area extracting step of extracting an object area from the input image, a feature point extracting step of extracting a feature point from the object area, and an input using the feature point as a reference. A normalization step of normalizing the image, a coordinate axis setting step of defining a coordinate axis based on the feature point in the normalized image, and a signature detection for detecting an electronic signature from the normalized image based on the coordinate axis. The method is a digital watermark detection method including steps.

The feature point extracting step is characterized by extracting a plurality of feature points.

Further, in the normalizing step, the size of the object is normalized.

Further, the normalizing step is characterized in that the angle of the object is normalized.

Further, the region extracting step is characterized by extracting a human face portion.

[0022]

BEST MODE FOR CARRYING OUT THE INVENTION The present invention will be specifically described below with reference to the drawings showing the embodiments thereof.

Embodiment 1. 1 is a flow chart showing a digital watermark embedding method according to a first embodiment of the present invention. In the figure, 10 is a step of inputting an image, 11
Is a step of extracting an object region, 12 is a step of extracting feature points, 13 is a step of setting coordinate axes, 14 is a step of embedding a signature, and 15 is a step of outputting an image.

Next, the operation will be described. Step 1
1 detects a predetermined object from the image input in step 10 and cuts out the area. Next, step 12 extracts a predetermined characteristic point from the area. In step 13, the coordinate axes are set on the image based on the feature points.

Here, when the object area is extracted from the input image in step 11, an object area having many characteristic points, for example, a human face portion is suitable. However, the present invention is not only applicable to face images, but can be applied to all objects that can obtain feature points.

FIG. 2 shows an example of coordinate axes set when an object is a face and feature points are eyes. In the figure, the midpoint between the right and left eyes is the origin, the x-axis is horizontal in the image, and y is vertical
A coordinate system having two coordinate axes called axes is set. Here, using a plurality of feature points (two are used in this case) as in this example is effective in ensuring the accuracy of the coordinate axes, and at the same time, as described in the second embodiment below. Mandatory when normalizing the image to. In addition,
In FIG. 2, the eye position lies on the x-axis, but it does not have to be. In that case also, the origin of the coordinates is set to the midpoint of both eyes.

Next, in step 14, the signature is embedded in the image based on the coordinate axes set in step 13. This embedding is performed, for example, in WO99 / 0 described in the prior art.
The method of 1980 may be used. Step 15 outputs this as a watermarked image.

Therefore, according to the first embodiment, the object area is extracted from the image, the feature points are further obtained from the extracted object area, and the signature is embedded based on the feature points. Can be accurately embedded in position.

Further, since a plurality of feature points are extracted in step 12, the coordinate axes can be set accurately in step 13.

Further, since the human face portion is detected as an object to be detected when the object area is extracted in step 11, (1) the human face image is included in many images and the application range is Wide, (2) Human face images always have feature points such as eyes, nose, and mouth, and coordinate axes for digital watermarking are easy to define. (3) Human visual characteristics of face are sensitive and malicious third parties. The effect is that it is difficult to tamper with.

In the first embodiment, as shown in FIG. 2, the coordinate axes are orthogonal such that they are horizontal x-axis and vertical y-axis, but in the present invention, the coordinate axes are orthogonal. It does not have to be orthogonal and does not have to be orthogonal, and the coordinate axes are other than the two axes of the x-axis and the y-axis, that is, one axis,
Even if there are three or more axes, any number of axes may be used depending on the object in which the digital watermark is embedded. In addition, in the first embodiment, as shown in FIG. 2, a plurality of feature points are extracted.
Of course, only one feature point may be extracted if only one feature point needs to be extracted. These are the same in any of the following second to fourth embodiments.

Embodiment 2. FIG. 3 is a flow chart showing a digital watermark detection method according to the second embodiment of the present invention.
In the figure, 20 is a step of inputting an image, 21 is a step of extracting an object area, 22 is a step of extracting feature points, 23 is a step of normalizing an object area, 24 is a step of setting coordinate axes, and 25 is a signature. Embedded image, 26 is a step of performing denormalization, 27 is a step of combining regions, and 28 is a step of outputting an image.

Next, the operation will be described. Step 21
And step 22 operate similarly to step 11 and step 12 in the first embodiment. Therefore, at the output of step 22, the object area containing the feature points is obtained. In the next step 23, this area is subjected to normalization processing such as reduction or enlargement to a prescribed size or adjustment in a prescribed direction. That is, in this step 23,
The normalization processing shown in, for example, FIGS.
An image of the detected object is output after being normalized in a certain size and in a certain direction.

Steps 24 and 25 are respectively
The operation is the same as steps 13 and 14 in the first embodiment. Therefore, in the output of step 25,
A normalized image with an embedded signature is obtained. Next, step 26 performs the inverse transformation of step 23 to restore the normalized image to its original size. Here, in order to return the image data excluding the signature data to the original values before and after the normalization, in step 23 and step 26,
For example, it is appropriate to use a method of reversible transformation such as wavelet analysis or Laplacian pyramid. In step 27, the denormalized image in which the signature is embedded is replaced with the image portion of the object region extracted in step 21, and in step 28, this combined image is output as a watermarked image.

Here, when the object area is extracted from the input image in step 21, an object area having many characteristic points, for example, a human face portion is suitable. In this case, as will be described later, in step 22, the eye position is detected as a feature point, in step 23, the face angle and size are normalized, and the signature embedding process in step 25 is performed from the eye position. The signature may be made with reference to the coordinate axes that are determined. However, the present invention is not only applicable to face images, but can be applied to all objects that can obtain feature points.

FIG. 4 is a diagram showing an example of the object area normalization processing in step 23 shown in FIG. In the example of FIG. 4, for example, when an object called a face is detected from the input image and feature points called both eyes are extracted, the direction connecting both eyes of the face is horizontal, so the direction of the object area of the face is changed. It shows a case where only the size of the object region of the face is normalized without performing the normalization process. That is, specifically, for example,
When the distance between the eyes of the face is A as shown in FIG. 4A, if the distance between the eyes of the face is normalized to be M, as shown in FIG. Shows that the object area of the face is multiplied by M / A and is normalized so that the distance between the eyes of the face becomes M.

FIG. 5 is a diagram showing another example of the object area normalization processing in step 23 shown in FIG. In the example of FIG. 5, as in the case of FIG. 4, for example, an object called a face is detected from the input image, feature points of both eyes are extracted, and the distance between both eyes of the face is A. The case where the normalization processing of the size shown in FIG. 4 is not performed but only the direction of the object area of the face is normalized is shown. That is, specifically, for example, as shown in FIG.
When the direction connecting the eyes of the face is inclined by the angle a with respect to the horizontal direction, the face is rotated by the angle a as the center between the eyes as shown in FIG. 5B. It is shown that the normalization is performed so that the direction connecting both eyes of the is horizontal. To normalize the angle, for example, affine transformation or the like may be used.

FIG. 6 is a diagram showing still another example of the normalization processing of the object area in step 23 shown in the figure. This Figure 6
In the example, the normalization processing for both the size and the direction, that is, the normalization processing in which FIGS. 4 and 5 are combined is shown. That is, in the example of FIG. 6, when an object called a face is extracted from the input image and both eyes are detected as feature points, the distance between the eyes of the face is A as shown in FIG. 6A.
If the direction connecting the eyes of the face is tilted by the angle a with respect to the horizontal direction, first, as shown in FIG.
As shown in (b), the angle a is centered between the eyes of the face.
Rotate and normalize so that the direction connecting the eyes of the face is horizontal. Next, when the distance between the eyes of the face is A, in order to normalize the distance between the eyes to be M, the object area of the face is multiplied by M / A, and finally As shown in FIG. 6D, the face is normalized so that the distance between the eyes is M and the direction connecting the eyes of the face is the horizontal direction.

On the contrary, an input image as shown in FIG. 6A is first subjected to a size normalization process by multiplying the object area of the face by M / A, and then the size of FIG. As shown in FIG. 6, the distance between the eyes of the face is normalized to be M, and then the face is rotated by an angle a about the distance between the eyes of the face, and finally, as shown in FIG. The distance between the eyes of the face may be M and the direction connecting the eyes of the face may be normalized so that the direction is horizontal. The normalized image of (d) may be obtained.

Therefore, according to the second embodiment, the object area is detected from the image, the feature points are extracted from the area, and the object area is normalized using the extracted feature points,
Since the signature is embedded in the normalized image, even if part of the image is lost or if geometric processing such as reduction, enlargement, or rotation is performed, the object area is extracted and normalized. Then, the embedding position of the signature can be easily recognized by previously recognizing even the normalization condition (in the above example, the distance between both eyes is M and the direction connecting both eyes is the horizontal direction). So, without matching with the original image,
There is an effect that alignment for watermark detection can be performed.

Further, since a plurality of feature points are extracted in step 22, the coordinate axes can be set accurately in step 23.

Further, since the human face portion is detected as the object to be detected when the object area is extracted in step 21, (1) the human face image is included in many images and the application range is Wide, (2) Human face images always have feature points such as eyes, nose, and mouth, and coordinate axes for digital watermarking are easy to define. (3) Human visual characteristics of face are sensitive and malicious third parties. The effect is that it is difficult to tamper with.

Embodiment 3. FIG. 7 is a flow chart showing a digital watermark detection method according to the third embodiment of the present invention.
In the figure, 30 is a step of inputting an image, 31 is a step of extracting an object region, 32 is a step of extracting feature points, 33 is a step of setting coordinate axes, and 34 is a step of detecting a signature.

Next, the operation will be described. Step 30
Steps 33 to 33 are the same as steps 10 to 13 of the first embodiment. Therefore, step 33
The output of is the coordinate axis set on the image of the detected object based on the feature points.

Next, in step 34, the embedded signature is detected from this image. As a method of this detection, for example, the method shown in FIG. 9 described in WO99 / 01980 described as a conventional example, that is, the wavelet transform process of step 90, the spread spectrum process of step 91, and the block division of step 92 are performed. Then, the Fourier transform processing of step 93 is further performed to check the value of the DC component of the Fourier transform.

As in the case of the first embodiment,
When extracting the object area from the input image in step 31, an object area having many feature points, for example, a part such as a human face is suitable, and in step 32, coordinate axes are determined based on the face feature points (eyes etc.) In the signature detection process of step 34, the signature may be detected with reference to the coordinate axes. Further, the present invention is not only applied to a face image, but can be applied to any object that can obtain a feature point.

Therefore, according to the third embodiment, since the object area is extracted from the image and the signature is detected based on the feature points, even if a part of the image is lost, the signature is detected. There is an effect that the embedding position of can be easily recognized and the position for watermark detection can be aligned without collating with the original image.

Fourth Embodiment FIG. 8 is a flow chart showing a digital watermark detection method according to the fourth embodiment of the present invention.
In the figure, 40 is a step of inputting an image, 41 is a step of extracting an object area, 42 is a step of extracting feature points, 43 is a step of normalizing an object area, 44 is a step of setting coordinate axes, and 45 is a signature. Is a step of detecting.

Next, the operation will be described. Step 40
Steps 44 to 44 are the same as steps 10 to 14 of the second embodiment. Therefore, step 43
The output of is obtained by normalizing an image of a detected object in a certain size and in a certain direction, and setting coordinate axes based on the feature points in the image.

Next, in step 45, the embedded signature is detected from this image. As a method of this detection, for example, the method shown in FIG. 9 described in WO99 / 01980 described as a conventional example, that is, the wavelet transform process of step 90, the spread spectrum process of step 91, and the block division of step 92 are performed. Then, the Fourier transform processing of step 93 is further performed to check the value of the DC component of the Fourier transform.

As in the case of the first embodiment,
When extracting the object area from the input image in step 41, an object area having many feature points, for example, a portion such as a human face is suitable. In step 42, the face feature points (eyes etc.) are extracted, After normalizing the object area, a coordinate axis is determined based on the feature points in step 44, and the signature detection processing in step 45 may be performed by detecting the signature based on this coordinate axis. Further, the present invention is
The object is not only applied to the face image, but can be applied to any object as long as the feature point can be obtained.

Therefore, according to the fourth embodiment, since the object area is extracted from the image and the signature is detected from the normalized image, part of the image is lost, or the image is reduced or enlarged. Even when geometrical processing such as rotation is performed, the normalization condition (distance M between both eyes in the above example, that the direction connecting both eyes is the horizontal direction) is recognized in advance. If so, the embedded position of the signature can be easily recognized, and the position for watermark detection can be aligned without collating with the original image.

[0053]

As described above, according to the present invention,
Since the object area is extracted from the image and the signature is embedded in the image based on the coordinates defined using the feature points,
Even if a part of the image is lost, there is an effect that alignment for watermark detection can be performed without collating with the original image.

Further, since the signature is detected from the image obtained by normalizing the extracted object area, even if the image is geometrically transformed, the watermark detection can be performed without collating with the original image. There is an effect that can be aligned.

Further, since the signature is detected from the image in which the size of the extracted object area is normalized, the watermark is detected without collating with the original image even when the image is enlarged or reduced. There is an effect that can be aligned.

Further, since the signature is detected from the image obtained by normalizing the direction of the extracted object area, even if the image is rotated, alignment for watermark detection is performed without collating with the original image. There is an effect that can be.

Further, since the embedding or the detection position is calculated by using a plurality of feature points, there is an effect that the positioning for watermark detection can be performed with high accuracy.

Further, in the digital watermark embedding method or detection method according to the present invention, since the human face part is particularly detected as the object to be detected, (1) the human face image is included in many images. Wide application range (2)
Human face images always have characteristic points such as eyes, nose, and mouth, and coordinate axes for digital watermarking are easy to determine. (3) Human visual characteristics for faces are sensitive and difficult to tamper with malicious third parties. Has the effect of.

[Brief description of drawings]

FIG. 1 is a flowchart showing an operation of a digital watermark embedding method according to a first embodiment of the present invention.

FIG. 2 is an explanatory diagram showing a normalizing operation of the present invention.

FIG. 3 is a flowchart showing the operation of the digital watermark embedding method according to the second embodiment of the present invention.

FIG. 4 is an explanatory diagram showing a normalizing operation of the present invention.

FIG. 2 is an explanatory diagram showing a normalizing operation of the present invention.

FIG. 6 is an explanatory diagram showing a normalizing operation of the present invention.

FIG. 7 is a flowchart showing the operation of the digital watermark detection method according to the first embodiment of the present invention.

FIG. 8 is a flowchart showing the operation of the digital watermark detection method according to the second embodiment of the present invention.

FIG. 9 is a flowchart showing the operation of a conventional digital watermark embedding method.

FIG. 10 is a flow chart showing the operation of a conventional digital watermark embedding method and its detection method.

[Explanation of symbols]

10, 20, 30, 40 Image input steps, 1
1, 21, 31, 41 Region extraction step, 42 Feature point extraction step, 23, 43 Normalization step, 44 Coordinate axis setting step, 14, 25, 8
4, 104 Steps for embedding signature, 34, 45, 10
9 signature detection step, 26 inverse normalization step, 27 synthesis step, 15, 28 image output step, 80 wavelet transform step, 8
1 Spread spectrum step, 82 block division step, 83 Fourier transform step, 100
Original image, 101, 106 object extraction step, 102, 107 area specifying step, 103,
108 block identification step, 105 watermark image.

   ─────────────────────────────────────────────────── ─── Continued front page    F term (reference) 5B057 CA19 CB19 CE08 CE09 CG07                       DA08 DC05 DC36                 5C053 FA13 GB06 LA06 LA11                 5C076 AA14 BA06                 5J104 AA14 KA16

Claims (12)

[Claims] 1. An input step of inputting an image, an area extracting step of extracting an object area from the input image,
A feature point extraction step of extracting a feature point from the object area, a coordinate axis setting step of determining a coordinate axis based on the feature point in the input image, and a digital signature embedded in the image input based on the coordinate axis A digital watermark embedding method comprising an embedding step and an output step of outputting the signed image. 2. An input step of inputting an image, an area extracting step of extracting an object area from the input image,
A feature point extracting step of extracting a feature point from the object area, a normalization step of normalizing the input image with the feature point as a reference, and a coordinate axis for defining a coordinate axis with the feature point as a reference in the normalized image. A setting step, a signature embedding step that embeds a digital signature in an image that is normalized based on its coordinate axis, a denormalization step that performs conversion that is the reverse of normalization to the image that has the signature embedded, and its denormalization A digital watermark embedding method comprising a combining step of combining the combined image with an original image, and an output step of outputting the combined image. 3. The feature point extracting step extracts a plurality of feature points.
The method for embedding a digital watermark described in. 4. The digital watermark embedding method according to claim 2, wherein the normalizing step normalizes the size of the object. 5. The digital watermark embedding method according to claim 2, wherein the normalizing step normalizes an angle of an object. 6. The digital watermark embedding method according to claim 1, wherein the region extracting step extracts a human face portion. 7. An input step of inputting an image, an area extracting step of extracting an object area from the input image,
A feature point extraction step of extracting a feature point from the object area, a coordinate axis setting step of determining a coordinate axis based on the feature point in the input image, and a digital signature is detected from the image input based on the coordinate axis. A digital watermark detection method including a signature detection step. 8. An input step of inputting an image, an area extracting step of extracting an object area from the input image,
A feature point extracting step of extracting a feature point from the object area, a normalization step of normalizing the input image with the feature point as a reference, and a coordinate axis for defining a coordinate axis with the feature point as a reference in the normalized image. A digital watermark detection method comprising a setting step and a signature detection step of detecting a digital signature from an image normalized based on its coordinate axes. 9. The feature point extracting step extracts a plurality of feature points.
The method for detecting a digital watermark described in. 10. The method of detecting a digital watermark according to claim 8, wherein the normalizing step normalizes the size of the object. 11. The digital watermark detection method according to claim 8, wherein the normalizing step normalizes an angle of an object. 12. The method according to claim 7, wherein the area extracting step extracts a human face portion.
The method for detecting a digital watermark described in.